The patch titled SLUB: Conform more to SLABs SLAB_HWCACHE_ALIGN behavior has been added to the -mm tree. Its filename is slub-core-conform-more-to-slabs-slab_hwcache_align-behavior.patch *** Remember to use Documentation/SubmitChecklist when testing your code *** See http://www.zip.com.au/~akpm/linux/patches/stuff/added-to-mm.txt to find out what to do about this ------------------------------------------------------ Subject: SLUB: Conform more to SLABs SLAB_HWCACHE_ALIGN behavior From: Christoph Lameter <clameter@xxxxxxx> Currently SLUB is using a strict L1_CACHE_BYTES alignment if SLAB_HWCACHE_ALIGN is specified. SLAB does not align to a cacheline if the object is smaller than half of a cacheline. Small objects are then aligned by SLAB to a fraction of a cacheline. Make SLUB just forget about the alignment requirement if the object size is less than L1_CACHE_BYTES. It seems that fractional alignments are no good because they grow the object and reduce the object density in a cache line needlessly causing additional cache line fetches. If we are already throwing the user suggestion of a cache line alignment away then lets do the best we can. Maybe SLAB_HWCACHE_ALIGN also needs to be tossed given its wishy-washy handling but doing so would require an audit of all kmem_cache_allocs throughout the kernel source. In any case one needs to explictly specify an alignment during kmem_cache_create to either slab allocator in order to ensure that the objects are cacheline aligned. [Patch has a nice memory compaction effect on 32 bit platforms] Signed-off-by: Christoph Lameter <clameter@xxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- mm/slub.c | 18 ++++++++++++++---- 1 file changed, 14 insertions(+), 4 deletions(-) diff -puN mm/slub.c~slub-core-conform-more-to-slabs-slab_hwcache_align-behavior mm/slub.c --- a/mm/slub.c~slub-core-conform-more-to-slabs-slab_hwcache_align-behavior +++ a/mm/slub.c @@ -1474,9 +1474,19 @@ static int calculate_order(int size) * various ways of specifying it. */ static unsigned long calculate_alignment(unsigned long flags, - unsigned long align) + unsigned long align, unsigned long size) { - if (flags & (SLAB_MUST_HWCACHE_ALIGN | SLAB_HWCACHE_ALIGN)) + /* + * If the user wants hardware cache aligned objects then + * follow that suggestion if the object is sufficiently + * large. + * + * The hardware cache alignment cannot override the + * specified alignment though. If that is greater + * then use it. + */ + if ((flags & (SLAB_MUST_HWCACHE_ALIGN | SLAB_HWCACHE_ALIGN)) && + size > L1_CACHE_BYTES / 2) return max_t(unsigned long, align, L1_CACHE_BYTES); if (align < ARCH_SLAB_MINALIGN) @@ -1664,7 +1674,7 @@ static int calculate_sizes(struct kmem_c * user specified (this is unecessarily complex due to the attempt * to be compatible with SLAB. Should be cleaned up some day). */ - align = calculate_alignment(flags, align); + align = calculate_alignment(flags, align, s->objsize); /* * SLUB stores one object immediately after another beginning from @@ -2252,7 +2262,7 @@ static struct kmem_cache *find_mergeable return NULL; size = ALIGN(size, sizeof(void *)); - align = calculate_alignment(flags, align); + align = calculate_alignment(flags, align, size); size = ALIGN(size, align); list_for_each(h, &slab_caches) { _ Patches currently in -mm which might be from clameter@xxxxxxx are slab-introduce-krealloc.patch slab-introduce-krealloc-fix.patch ia64-sn-xpc-convert-to-use-kthread-api-fix.patch add-apply_to_page_range-which-applies-a-function-to-a-pte-range.patch safer-nr_node_ids-and-nr_node_ids-determination-and-initial.patch use-zvc-counters-to-establish-exact-size-of-dirtyable-pages.patch slab-ensure-cache_alloc_refill-terminates.patch smaps-extract-pmd-walker-from-smaps-code.patch smaps-add-pages-referenced-count-to-smaps.patch smaps-add-clear_refs-file-to-clear-reference.patch smaps-add-clear_refs-file-to-clear-reference-fix.patch smaps-add-clear_refs-file-to-clear-reference-fix-fix.patch slab-use-num_possible_cpus-in-enable_cpucache.patch extend-print_symbol-capability-fix.patch extend-print_symbol-capability-fix-fix.patch i386-use-page-allocator-to-allocate-thread_info-structure.patch slub-core.patch slub-fix-numa-bootstrap.patch slub-use-correct-flags-to-check-for-dma-cache.patch slub-treat-slab_hwcache_align-as-a-mininum-and-not-as-the-alignment.patch slub-core-minor-fixes.patch slub-core-use-enum-for-tracking-modes-instead-of-integers.patch slub-core-fix-another-numa-bootstrap-issue.patch slub-core-fix-object-counting.patch slub-core-drop-version-number.patch slub-core-tidy.patch slub-core-tidy-2.patch slub-core-tidy-3.patch slub-core-tidy-4.patch slub-core-tidy-5.patch slub-core-tidy-6.patch slub-core-tidy-7.patch slub-core-tidy-8.patch slub-core-tidy-9.patch slub-core-we-do-not-need-ifdef-config_smp-around-bit-spinlocks.patch slub-core-printk-facility-level-cleanup.patch slub-core-kmem_cache_close-is-static-and-should-not-be-exported.patch slub-core-add-explanation-for-defrag_ratio-=-100.patch slub-core-add-explanation-for-locking.patch slub-core-add-explanation-for-locking-fix.patch slub-core-explain-the-64k-limits.patch slub-core-explain-sizing-of-slabs-in-detail.patch slub-core-explain-sizing-of-slabs-in-detail-fix.patch slub-core-add-checks-for-interrupts-disabled.patch slub-core-use-__print_symbol-instead-of-kallsyms_lookup.patch slub-core-missing-inlines-and-statics.patch slub-fix-cpu-slab-flushing-behavior-so-that-counters-match.patch slub-extract-finish_bootstrap-function-for-clean-sysfs-boot.patch slub-core-fix-kmem_cache_destroy.patch slub-core-fix-validation.patch slub-core-add-after-object-padding.patch slub-core-resiliency-fixups.patch slub-core-resiliency-fixups-fix.patch slub-core-resiliency-test.patch slub-core-update-cpu-after-new_slab.patch slub-core-fix-sysfs-directory-handling.patch slub-core-conform-more-to-slabs-slab_hwcache_align-behavior.patch slub-core-reduce-the-order-of-allocations-to-avoid-fragmentation.patch make-page-private-usable-in-compound-pages-v1.patch make-page-private-usable-in-compound-pages-v1-hugetlb-fix.patch optimize-compound_head-by-avoiding-a-shared-page.patch add-virt_to_head_page-and-consolidate-code-in-slab-and-slub.patch slub-fix-object-tracking.patch slub-enable-tracking-of-full-slabs.patch slub-enable-tracking-of-full-slabs-fix.patch slub-enable-tracking-of-full-slabs-add-checks-for-interrupts-disabled.patch slub-validation-of-slabs-metadata-and-guard-zones.patch slub-validation-of-slabs-metadata-and-guard-zones-fix-pageerror-checks-during-validation.patch slub-validation-of-slabs-metadata-and-guard-zones-remove-duplicate-vm_bug_on.patch slub-add-min_partial.patch slub-add-ability-to-list-alloc--free-callers-per-slab.patch slub-add-ability-to-list-alloc--free-callers-per-slab-tidy.patch slub-free-slabs-and-sort-partial-slab-lists-in-kmem_cache_shrink.patch slub-remove-object-activities-out-of-checking-functions.patch slub-user-documentation.patch slub-user-documentation-fix.patch slub-add-slabinfo-tool.patch slub-add-slabinfo-tool-update-slabinfoc.patch slub-major-slabinfo-update.patch slub-exploit-page-mobility-to-increase-allocation-order.patch slub-i386-support.patch slub-mm-only-make-slub-the-default-slab-allocator.patch quicklists-for-page-table-pages.patch quicklists-for-page-table-pages-avoid-useless-virt_to_page-conversion.patch quicklists-for-page-table-pages-avoid-useless-virt_to_page-conversion-fix.patch quicklist-support-for-ia64.patch quicklist-support-for-x86_64.patch quicklist-support-for-sparc64.patch slab-allocators-remove-obsolete-slab_must_hwcache_align.patch kmem_cache-simplify-slab-cache-creation.patch slab-allocators-remove-slab_debug_initial-flag.patch slab-allocators-remove-slab_debug_initial-flag-locks-fix.patch slab-allocators-remove-multiple-alignment-specifications.patch slab-allocators-remove-slab_ctor_atomic.patch fault-injection-fix-failslab-with-config_numa.patch mm-fix-handling-of-panic_on_oom-when-cpusets-are-in-use.patch slab-shutdown-cache_reaper-when-cpu-goes-down.patch mm-implement-swap-prefetching.patch revoke-core-code-slab-allocators-remove-slab_debug_initial-flag-revoke.patch readahead-state-based-method-aging-accounting.patch - To unsubscribe from this list: send the line "unsubscribe mm-commits" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html